Globally Consistent RGB-D SLAM with 2D Gaussian Splatting

2025-06-01

AI Summary

To address geometric distortion and map drift in RGB-D SLAM caused by depth rendering inconsistency and inefficient loop closure detection in 3D Gaussian splatting-based online globally consistent mapping, this paper proposes the first real-time SLAM framework leveraging 2D Gaussian rasterization. Our method replaces volumetric 3D Gaussians with lightweight, differentiable 2D Gaussians as the map representation, significantly improving rendering efficiency and depth consistency. We introduce a depth-consistency rendering loss jointly optimized with camera pose estimation. For robust loop closure and relocalization, we integrate the MASt3R foundation model for efficient 3D scene understanding. Furthermore, we adopt a local active map mechanism coupled with joint photometric-geometric optimization. Extensive experiments demonstrate that our approach achieves superior tracking accuracy, surface reconstruction quality, and global consistency compared to state-of-the-art rendering-based SLAM methods, while maintaining high-fidelity rendering and real-time performance.

Technology Category

• Intelligent Robots: Localization, Mapping, and Navigation
• Computer Vision: 3D Computer Vision
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Recently, 3D Gaussian splatting-based RGB-D SLAM displays remarkable performance of high-fidelity 3D reconstruction. However, the lack of depth rendering consistency and efficient loop closure limits the quality of its geometric reconstructions and its ability to perform globally consistent mapping online. In this paper, we present 2DGS-SLAM, an RGB-D SLAM system using 2D Gaussian splatting as the map representation. By leveraging the depth-consistent rendering property of the 2D variant, we propose an accurate camera pose optimization method and achieve geometrically accurate 3D reconstruction. In addition, we implement efficient loop detection and camera relocalization by leveraging MASt3R, a 3D foundation model, and achieve efficient map updates by maintaining a local active map. Experiments show that our 2DGS-SLAM approach achieves superior tracking accuracy, higher surface reconstruction quality, and more consistent global map reconstruction compared to existing rendering-based SLAM methods, while maintaining high-fidelity image rendering and improved computational efficiency.

Problem

• Improving depth rendering consistency in RGB-D SLAM
• Enabling efficient online loop closure for global mapping
• Achieving accurate 3D reconstruction with optimized camera poses

Innovation

• Uses 2D Gaussian splatting for map representation
• Implements depth-consistent camera pose optimization
• Leverages MASt3R for efficient loop detection